의료 빅데이터 - 개념 입문과 임상의사가 할 수 있는 일

Big Data Driven Healthcare 충북대학교병원 내분비내과 최형진

Contents

1. What is Big Data?

2. Healthcare Big Data

① Electrical Health Records (EHR) Structured/Unstructured Data

② Medical Images

③ National Healthcare Data

④ Behavior/Sensor Data

⑤ Genetic Data

3. Clinical and Research Applications

What is Big Data?

미국 내분비 학회

미국 골대사 학회

Big Data Dimensions

Big Meal?

Volume Variety

Velocity

(중환자실 심전도)

Data Variety

Architecture of Big Data Analytics

2014 Big data analytics in healthcare- promise and potential

Machine Learning

2014 Big data bioinformatics

Tipping Point for Big Data Healthcare

2013 McKinsey The big data revolution in healthcare

Gartner's Hype Cycles

Hypothesis Driven Science Data Driven Science

Hypothesis

Collect

Data

Data

Generate

Hypothesis

Analyze

Analyze

Candidate Gene

Approach

Genome-wide

Approach

Choose a Gene

from Prior Knowledge

Analyze the Gene

Analyze ALL Genes

Discover Novel Findings

GWAS

(Genome wide association study)

SNP chip

Whole Genome

SNP Profiling (500K~1M SNPs)

Common Variant

Estrada et al., Nature Genetics, 2012

+ novel targets

for bone biology

Recent largest GWAS

GEFOS consortium

2010 An Environment-Wide

Association Study (EWAS) on Type

2 Diabetes Mellitus

Environment-Wide Association Study (EWAS)

Contents

1. What is Big Data?

2. Healthcare Big Data

① Electrical Health Records (EHR) Structured/Unstructured Data

② Medical Images

③ National Healthcare Data

④ Behavior/Sensor Data

⑤ Genetic Data

3. Clinical and Research Applications

Genomics

Wearable/

Smart Device

Medical Informatics

Genome Sequencing

Quantified Self

Health Avatar

Electronic Medical Records (EMR)

Medical Images (MRI, CT)

National Healthcare data (Insurance)

Contents

1. What is Big Data?

2. Healthcare Big Data

① Electrical Health Records (EHR) Structured/Unstructured Data

② Medical Images

③ National Healthcare Data

④ Behavior/Sensor Data

⑤ Genetic Data

3. Clinical and Research Applications

Electronic Health Records

2012 NRG Mining electronic health records- towards better research applications and clinical care

Common EHR Data

Joshua C. Denny Chapter 13: Mining Electronic Health Records in the Genomics Era. PLoS Comput Biol. 2012 December; 8(12):

International Classification of Diseases (ICD)

Current Procedural Terminology (CPT)

Medication Data

Lab Data

Big Data Analysis

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.81/Creatinine

한 환자의 10년간 신장기능의 변화 전체 환자의 당 조절 정도 분포

충북대 정보통계학과

허태영 교수

PCA Analysis 혈당

신장기능

Clinical Notes

밤동안 저혈당수면 Lt.foot rolling Keep떨림,

식은땀, 현기증, 공복감, 두통, 피로감등의 저혈당 에 저혈당 이 있을 즉알려주도록 밤사이 특이호소 수면유지상처와 통증 상처부위 출혈

oozing, severe pain 알리도록 고혈당 처방된 당뇨식이의 중요성과 간식을 자제하도록 .고혈당 ,,관리 방법 .당뇨약 이해 잘 하고 수술부위

oozing Rt.foot rolling keep드레싱 상태를 고혈당 고혈당 의식변화 BST 387 checked.고혈당으로 인한 구강 내 감염 위해 식후 양치, gargle

등 구강 위생 격려.당뇨환자의 발관리 방법에 .

목표 혈당, 목표 당화혈색소에 .식사를 거르거나 지연하지 않도록 .식사요법, 운동요법, 약물요법을 정확히 지키는 것이 중요을 .처방된 당뇨식이의 중요성과 간식을 자제하도록 .고혈당 ,,관리 방법 .혈당 정상 범위임rt foot rolling

중으로 pain호소 밤사이 수면양호걱정신경 예민감정변화 중임감정을 표현하도록 지지하고

경청기분상태 condition 조금 나은 듯 하다고 혈당 조절과 관련하여 신경쓰는 모습 보이며 혈당

self로 측정하는 모습 보임혈당 조절에 안내하고 불편감 지속알리도록고혈당 고혈당 의식변화 고혈당 허약감 지남력 혈당조절 안됨고혈당으로 인한 구강 내 감염 위해 식후 양치, gargle

등 구강 위생 격려.당뇨환자의 정기점검 내용과

빈도에 .BST 140 으로 저혈당 호소 밤동안 저혈당수면 Lt.foot rolling Keep떨림, 식은땀, 현기증, 공복감, 두통, 피로감등의 저혈당 에 저혈당 이 있을 즉알려주도록 pain 및 불편감 호소 WA 잘고혈당 고혈당 의식변화 고혈당 허약감 지남력 혈당조절 안됨식사요법, 운동요법,

약물요법을 정확히 지키는 것이 중요을 .저혈당/고혈당 과 대처법에 .혈당정상화, 표준체중의

유지, 정상 혈중지질의 유지에 .고혈당 ,,관리

방법 .혈당측정법,인슐린 자가 투여법, 경구투약,수분 섭취량,대체 탄수화물,의료진의 도움이

필요한 사항에 교혈당 정상 범위임수술부위

oozing Rt.foot rolling keep수술 부위 (출혈, 통증, 부종)수술부위 출혈 상처부위 oozing

Wound 당겨지지 않도록 적절한 체위 취하기

설명감염 발생 위험 요인 수술부위 출혈 밤동안

저혈당 호소 수면 양상 양호rt foot rolling유지하며 감염이나 심한 통증등 침상안정중임BST

420 checkd. R1 알림 obs 하자Rt DM foot site

pain oozing 없는 상태로 저혈당 은 낮동안에는 . BST BST 347 mg /dl checked. R1notify 후

apidra 10U sc 주사.안전간호 시행안전간호 시행

간호기록지 Word Cloud

Natural Language Processing (NLP)

Contents

1. What is Big Data?

2. Healthcare Big Data

① Electrical Health Records (EHR) Structured/Unstructured Data

② Medical Images

③ National Healthcare Data

④ Behavior/Sensor Data

⑤ Genetic Data

3. Clinical and Research Applications

Medical

Images

Brain Vasculature

Color =

Vessel Diameters 2011 Optimization of MicroCT Imaging and Blood Vessel Diameter Quantitation of Preclinical Specimen Vasculature with Radiopaque Polymer Injection Medium

fMRI analysis

2013 Sciencce Functional interactions as big data in the human brain

Functional Connectivity

2013 Sciencce Functional interactions as big data in the human brain

N=50,000 voxels

N(N – 1)/2 = 1,249,975,000 Pairs

Seed Region Based Analysis

비만 fMRI 연구

tDCS stimulation

2mA for 20min

Behavioral task 1

(food image ratings)

~7min

3

+

Food presentation

max 4sec

Feedback

1sec

Fixation

1~10sec

Blood test/survey

fMRI

Tim

e

Blood test/survey

Resting MRI

5-6min

T1 structure MRI

5-6min

Behavior task 2

(food intake rating)

Bone Quality and

TBS (Trabecular Bone Score)

Clinical Implication of TBS

2014 Endocrine. Utility of the trabecular bone score (TBS) in secondary osteoporosis

Contents

1. What is Big Data?

2. Healthcare Big Data

① Electrical Health Records (EHR) Structured/Unstructured Data

② Medical Images

③ National Healthcare Data

④ Behavior/Sensor Data

⑤ Genetic Data

3. Clinical and Research Applications

Overview of secondary data in

public health by data source

Data Variety

Anti-hypertensive

prescriptions

(2008-2011)

N = 8,315,709

New users

N = 2,357,908

Age ≥ 50 yrs

Monotherapy

Compliant user (MPR≥80%)

No previous fracture

N = 528,522

Prevalent users

N = 5,957,801

Excluded

Age <50

Combination therapy

Inadequate compliance

Previous fracture

N = 1,829,386

Final study population

심평원 빅데이터 연구

고혈압약과 골절

Choi et al., in sumission

Design

BB

CCB ACEI

2007 2011

Hypertension

Fracture Incidence

Fracture Incidence

Cohort study (Health Insurance Review & Assessment Service)

New-user design (drug-related toxicity)

Non-user comparator (hypertension without medication)

Fracture Incidence

2008

Fracture

Fracture Fracture

* Exclude

- Fractures

- Medications (New User)

Fracture

No

Fracture

BB

BB

BB

BB

BB

ARB

ARB

ARB ARB

ARB

BB

Group

ARB

Group

First Prescription of Antihypertensives

Fracture Outcomes

2007 2011.1. 2008 2009 2010 2011.12.31.

Exclusion of Fracture

within 6 months

Distribution of ARB MPR

(Histogram)

ARB Non-user

20

Fre

qu

en

cy D

en

sity

ARB user

80 120

Medication Possession Ratio (MPR)

Total prescription days

Observation days

350 days (Prescription)

365 days (Observation)

MPR

96%

MPR (%)

Fracture rates per 10,000 person-years (95% CI)

819

Fracture Rates (per 10,000 Person-Years)

Total

Male

Female

AB: alpha-adrenergic blocker

ACEI: angiotensin converting enzyme inhibitor

DIUR: diuretics

CCB: calcium channel blocker

BB: beta-adrenergic blocker

ARB: angiotesin-receptor blocker

Contents

1. What is Big Data?

2. Healthcare Big Data

① Electrical Health Records (EHR) Structured/Unstructured Data

② Medical Images

③ National Healthcare Data

④ Behavior/Sensor Data

⑤ Genetic Data

3. Clinical and Research Applications

혈당관리

혈당 관리

혈당관리

Apple Healthbook

Contents

1. What is Big Data?

2. Healthcare Big Data

① Electrical Health Records (EHR) Structured/Unstructured Data

② Medical Images

③ National Healthcare Data

④ Behavior/Sensor Data

⑤ Genetic Data

3. Clinical and Research Applications

DNA

mRNA

Protein

Metabolite

Epigenetics

Genetics Information and OMICs

Genomics

Epigenomics

Transcriptomics

Proteomics

Metabolomics

Disease genetic susceptibility Cancer driver

somatic mutation

Pharmacogenomics

Targeted

Cancer Treatment

(EGFR)

Causal

Variant

Targeted Drug

(MODY-SU)

Drug Efficacy/Side Effect

Related Genotype

(CYP, HLA)

Genetic Diagnosis

(Mendelian,

Cystic fibrosis) Molecular

Classification

- Prognosis

(Leukemia)

Hereditary

Cancer

(BRCA)

Microbiome

(Bacteria,

Virus)

Genomic Medicine

Risk prediction

(Complex disease,

Diabetes)

Germline Variants

저의 유전자

분석 결과를

반영하여 진료해주세요!! 헠?

Voxel-wise GWAS

2014 Nature Neuroscience. Whole-genome analyses of whole-brain data- working within an expanded search space

Connectome-wide GWAS

2014 Nature Neuroscience. Whole-genome analyses of whole-brain data- working within an expanded search space

Disease GWAS vs. Whole-brain GWAS

2014 Nature Neuroscience. Whole-genome analyses of whole-brain data- working within an expanded search space

2014 JAMA Finding the Missing Link for Big Biomedical Data

Contents

1. What is Big Data?

2. Healthcare Big Data

① Electrical Health Records (EHR) Structured/Unstructured Data

② Medical Images

③ Government Data

④ Behavior/Sensor Data

⑤ Genetic Data

3. Clinical and Research Applications

발전한다?

New Value Pathways

Individual Decision Making

Medical Big Data

Artificial Intelligence

Jeopardy!

2011년 인간 챔피언 두 명 과 퀴즈 대결을 벌여서 압도적인 우승을 차지

2013 PLOS CB Reassessing Google Flu Trends Data for Detection of Seasonal and Pandemic Influenza

Google Flu Trends

Social Network and

Obesity Prevalence

2013 PLOS One. Assessing the Online Social Environment for Surveillance of Obesity Prevalence

2014 Science. Big data. The parable of Google Flu- traps in big data analysis

2014 Science. Big data. The parable of Google Flu- traps in big data analysis

Big data platform model by Korea Institute of

Drug Safety and Risk Management

Medical Big Data

Medical Big Data

Medical Big Data

R vs. Stata vs. SAS

빅데이터 연구 적용

전통적인 관점 연구

Large scale

(unstructured)

data

Summary

(Modify)

Classical hypothesis driven study

새로운 관점 연구

Hypothesis Generating Study